A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

VibeVoice ASR Streaming

Transcribe authorized WAV files locally with the 1.5B streaming checkpoint. Native speaker-prefixed text and complete chunk outputs arrive progressively; hotword context is optional.

Desktop inferencePrivate · Runs on your computer
About this model

Transcribe authorized WAV files locally with the 1.5B streaming checkpoint. Native speaker-prefixed text and complete chunk outputs arrive progressively; hotword context is optional.

Created by
Microsoft
License
MIT
Model released
Sep 3, 2026
Source repository created
Sep 2, 2026
Model weights · BF16
5.6 GB

Best for

  • Local speaker-attributed file transcription with manual review
  • Incremental model-chunk ASR experiments on authorized recordings

Keep in mind

  • Speech, names, punctuation and speaker assignments can be wrong, omitted, repeated or hallucinated. Speaker labels are predictions, not identity proof; review outputs before consequential use.
  • Upstream lists Chinese,English,French,German,Italian,Japanese,Korean,Portuguese,Russian,Spanish. Local quality evidence covers English only, not all ten languages.
  • This adapter accepts a complete mono/stereo RIFF WAV up to120seconds and200MB,8–384kHz; decodes/downmixes/resamples to24kHz on CPU. It does not feed a live microphone or certify hour-long sessions.
  • Chunk size22frames and4frame lookahead are fixed by the checkpoint: about2.933s and0.533s. Chunk windows are not word or speaker-turn timestamps.
  • Raw native text retains Speaker prefixes or explicit [Silence] annotations. Complete chunk text is preserved without fabricated JSON records or timestamps.
  • Streaming callbacks are provisional cumulative snapshots. Replace text/chunks on each callback; CPU retry attempt2 replaces a failed GPU prefix. Final result is authoritative.
  • 7B/1.5B name decoder sizes; complete checkpoints include additional speech encoders. Stored BF16 weight bytes are separate from measured execution dtype and full runtime/download/memory budgets.
  • Precision/device support is scoped to actual quality/performance-passing configurations on M3Max/macOS27.2; slow/unsupported candidates remain in the report. Other Mac/Windows hardware unverified.
  • Original Transformers5.22.1 maps these checkpoints to a TTS class with no ASR streaming method. Adapter uses reviewed pinned Microsoft ASR extension with Transformers4.57.6, no trust_remote_code or model repository scripts.
  • WebAI browser runtime/export is not verified; no browser inference advertised.

Processed on your deviceYour input is processed locally in the desktop app. Model and runtime files download from external hosts.

Target platforms · verification required

  • macos · aarch64, x86_64
  • windows · x86_64

Recommended configuration

BF16 · mps (verified platform only)

Sources & technical details
Original model
https://huggingface.co/microsoft/VibeVoice-ASR-Streaming-1.5B
Release verification
Verified
Release evidence
Official September3 streaming release links the Microsoft collection containing exact7B and1.5B checkpoints. This 1.5B streaming checkpoint is distinct from January21 offline ASR. Repository creationSeptember2 precedes official public release; update date is not used as release date.
Source updated
Sep 3, 2026
Native model files
https://huggingface.co/microsoft/VibeVoice-ASR-Streaming-1.5B
Files updated
Sep 3, 2026
Artifact revision
4262d23d8a539a6530cf64fbd0b1751ef9a30853

Desktop app

Sample result

VibeVoice ASR Streaming · 1.5B

Run VibeVoice ASR Streaming on your computer

Download the desktop app to load this model, provide input and view results locally.

Public desktop downloads are coming soon.

Device requirements

macOS Apple SiliconCPU / FP32Tested24.0 GiBRAM estimate
macOS Apple SiliconMPS / FP16Tested24.0 GiBRAM estimate
macOS Apple SiliconMPS / FP32Tested24.0 GiBRAM estimate
macOS Apple SiliconMPS / BF16Tested24.0 GiBRAM estimate
Windows Intel / AMDCPU / FP32Unverified24.0 GiBRAM estimate
macOS Intel / AMDCPU / FP32Unverified24.0 GiBRAM estimate
If your device falls short

MemoryAdvisory only: close other apps, reduce image/tile/token/batch settings or choose a smaller verified model/precision. CPU can need more RAM than GPU.

GPUIf the selected GPU runtime fails, existing CPU fallback may recover when enabled, but CPU has its own memory/runtime needs.

PlatformUse a compatible OS/architecture/runtime or another model. These estimates do not add hardware detection or block runs.

Benchmark evidence

Saved desktop measurements apply to the recorded hardware and runtime. Fallback cases report the device that actually ran; they do not prove GPU support on the requested device.

VibeVoice ASR Streaming · 1.5B

55 configurations tested
Passed55
Failed0
TestedNov 3, 2026
Runtimevibevoice-asr 1.0.0+16fb2cb

Quality criteria

No explicit quality gate recorded. Review per-case evidence; full certification is pending.

PrecisionDeviceResultPlatformInference / loadTask metric
BF16mpsAcceleratedPassedFastmacos · aarch641724 ms / 0 ms0.164 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641735 ms / 0 ms0.165 RTF
FP16mpsAcceleratedPassedFastmacos · aarch641738 ms / 0 ms0.165 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641756 ms / 0 ms0.167 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641766 ms / 0 ms0.167 RTF
FP16mpsAcceleratedPassedFastmacos · aarch641762 ms / 0 ms0.168 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641782 ms / 0 ms0.171 RTF
FP16mpsAcceleratedPassedFastmacos · aarch641809 ms / 0 ms0.173 RTF
FP16mpsAcceleratedPassedFastmacos · aarch641818 ms / 0 ms0.173 RTF
FP16mpsAcceleratedPassedFastmacos · aarch641834 ms / 0 ms0.175 RTF
BF16mpsAcceleratedPassedFastmacos · aarch646619 ms / 3212 ms0.189 RTF
FP16mpsAcceleratedPassedFastmacos · aarch646774 ms / 3558 ms0.193 RTF
FP16mpsAcceleratedPassedFastmacos · aarch642058 ms / 3211 ms0.195 RTF
FP16mpsAcceleratedPassedFastmacos · aarch642062 ms / 3410 ms0.195 RTF
BF16mpsAcceleratedPassedFastmacos · aarch642063 ms / 3237 ms0.196 RTF
BF16mpsAcceleratedPassedFastmacos · aarch642077 ms / 3194 ms0.197 RTF
BF16mpsAcceleratedPassedFastmacos · aarch64344 ms / 0 ms0.200 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642157 ms / 0 ms0.203 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642157 ms / 0 ms0.204 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642170 ms / 0 ms0.204 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642182 ms / 0 ms0.206 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642287 ms / 0 ms0.216 RTF
FP16mpsAcceleratedPassedFastmacos · aarch64373 ms / 0 ms0.220 RTF
FP16mpsAcceleratedPassedFastmacos · aarch642339 ms / 3510 ms0.222 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642463 ms / 3656 ms0.232 RTF
FP32mpsAcceleratedPassedFastmacos · aarch648189 ms / 3640 ms0.232 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642493 ms / 3713 ms0.235 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64417 ms / 0 ms0.241 RTF
FP16mpsAcceleratedPassedModeratemacos · aarch64833 ms / 0 ms0.262 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch64846 ms / 0 ms0.267 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641097 ms / 0 ms0.342 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch643826 ms / 3131 ms0.357 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643987 ms / 4345 ms0.372 RTF
FP32cpuPassedModeratemacos · aarch646031 ms / 0 ms0.556 RTF
FP32cpuPassedModeratemacos · aarch646139 ms / 0 ms0.566 RTF
FP32cpuPassedModeratemacos · aarch646201 ms / 0 ms0.571 RTF
FP32cpuPassedModeratemacos · aarch646235 ms / 0 ms0.574 RTF
FP32cpuPassedModeratemacos · aarch646281 ms / 0 ms0.579 RTF
FP32cpuPassedModeratemacos · aarch646447 ms / 3424 ms0.594 RTF
FP32cpuPassedModeratemacos · aarch6421199 ms / 10084 ms0.597 RTF
FP32cpuPassedModeratemacos · aarch646526 ms / 3499 ms0.602 RTF
FP32cpuPassedModeratemacos · aarch646566 ms / 3522 ms0.606 RTF
FP32cpuPassedModeratemacos · aarch647688 ms / 2924 ms0.708 RTF
FP32cpuPassedModeratemacos · aarch647740 ms / 0 ms0.712 RTF
FP32cpuPassedModeratemacos · aarch647758 ms / 0 ms0.713 RTF
FP32cpuPassedModeratemacos · aarch647837 ms / 0 ms0.720 RTF
FP32cpuPassedModeratemacos · aarch647836 ms / 2845 ms0.721 RTF
FP32cpuPassedModeratemacos · aarch647858 ms / 0 ms0.722 RTF
FP32cpuPassedModeratemacos · aarch648008 ms / 2677 ms0.736 RTF
FP32cpuPassedModeratemacos · aarch648021 ms / 0 ms0.737 RTF
FP32cpuPassedModeratemacos · aarch641383 ms / 0 ms0.740 RTF
FP32cpuPassedModeratemacos · aarch6426995 ms / 2960 ms0.760 RTF
FP32cpuPassedModeratemacos · aarch641715 ms / 0 ms0.887 RTF
FP32cpuPassedModeratemacos · aarch643075 ms / 0 ms0.921 RTF
FP32cpuPassedModeratemacos · aarch643734 ms / 0 ms1.118 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-03T00:06:54.156445+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "Reviewed Microsoft16fb2cb ASR / Transformers4.57.6 / Torch2.10.0", "interpreter": "App-ownedCPython3.13.18 private development venv; packaged managed provisioning tested separately" }
Runtime
vibevoice-asr 1.0.0+16fb2cb
Artifact revision
4262d23d8a539a6530cf64fbd0b1751ef9a30853
Fixture
{
  "file": "model-samples/desktop/granite-speech5/granite-speech5-jfk.wav",
  "size": 3880888,
  "sha1": "6fb9dfd0ec18bb61a6af995c0df0cd32ee2afe34",
  "sha256": "c72e27f68334ae37e037ef819fa5aa1fc1552aa840b93e308df4ffe52317bedb",
  "url": "https://assets.localai.tools/samples/v1/files/c72e27f68334ae37e037ef819fa5aa1fc1552aa840b93e308df4ffe52317bedb/granite-speech5-jfk.wav"
}
Rerun when
Exact checkpoint/tokenizer/preprocessor/reviewed runtime/framework/torch dependency changes; Accepted input decode/resample, chunking/KV/greedy generation or native output changes; Fixture/reference or speech projection/quality/performance gate changes; Relevant OS/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3499.407583993161,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6622.400374995777,
      "inferenceMs": 6525.607500021579,
      "firstChunkMs": 2363.594374997774,
      "audioSeconds": 11,
      "realTimeFactor": 0.6020363977268889,
      "audioSecondsPerSecond": 1.6856668133907264,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 10121.807958988938,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " The quick brown fox jumps over the lazy dog."
      },
      "plainTranscript": "The quick brown fox jumps over the lazy dog.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 3131.3904999988154,
      "inferenceMs": 3074.5081669883803,
      "firstChunkMs": 2145.2156669984106,
      "audioSeconds": 3.4,
      "realTimeFactor": 0.9209972058820045,
      "audioSecondsPerSecond": 1.1058679357259453,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 2,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 3131.3904999988154,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 0,
      "expectedSpeakerCount": 0,
      "speakerContent": {},
      "plainTranscript": "",
      "maximumConsecutiveRepeatedWords": 0,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1479.1136660205666,
      "inferenceMs": 1383.3208339929115,
      "firstChunkMs": 1479.1130000085104,
      "audioSeconds": 2,
      "realTimeFactor": 0.7395568330102833,
      "audioSecondsPerSecond": 1.44579619626422,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 1,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 1479.1136660205666,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6114.938666985836,
      "inferenceMs": 6031.323914969107,
      "firstChunkMs": 1972.0030419994146,
      "audioSeconds": 11,
      "realTimeFactor": 0.5559035151805305,
      "audioSecondsPerSecond": 1.8238118454721302,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 6114.938666985836,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6370.299333008006,
      "inferenceMs": 6280.85429102066,
      "firstChunkMs": 2075.988166994648,
      "audioSeconds": 11,
      "realTimeFactor": 0.579118121182546,
      "audioSecondsPerSecond": 1.7513541136794089,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 6370.299333008006,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6279.902915994171,
      "inferenceMs": 6201.450332009699,
      "firstChunkMs": 2008.9117079915013,
      "audioSeconds": 11,
      "realTimeFactor": 0.5709002650903792,
      "audioSecondsPerSecond": 1.7737786180793678,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 6279.902915994171,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6230.100917018717,
      "inferenceMs": 6139.218623022316,
      "firstChunkMs": 2074.0997090179008,
      "audioSeconds": 11,
      "realTimeFactor": 0.5663728106380652,
      "audioSecondsPerSecond": 1.791758964039749,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 6230.100917018717,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6316.436500026612,
      "inferenceMs": 6234.626377030509,
      "firstChunkMs": 2003.696083003888,
      "audioSeconds": 11,
      "realTimeFactor": 0.5742215000024192,
      "audioSecondsPerSecond": 1.7643398873950153,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 6316.436500026612,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12870434816,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3424.4031660200562,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6539.159666019259,
      "inferenceMs": 6447.085124993464,
      "firstChunkMs": 2167.328791023465,
      "audioSeconds": 11,
      "realTimeFactor": 0.5944690605472054,
      "audioSecondsPerSecond": 1.7061974189477065,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 9963.562832039315,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12051922944,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3521.8859169981442,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6661.469334008871,
      "inferenceMs": 6566.130625928054,
      "firstChunkMs": 2288.180042000022,
      "audioSeconds": 11,
      "realTimeFactor": 0.6055881212735337,
      "audioSecondsPerSecond": 1.6752636562793426,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 10183.355251007015,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12702449664,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 2845.3226670098957,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 7935.76849999954,
      "inferenceMs": 7835.537834063871,
      "firstChunkMs": 2558.038040995598,
      "audioSeconds": 11,
      "realTimeFactor": 0.7214334999999582,
      "audioSecondsPerSecond": 1.403860236904107,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 10781.091167009436,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " The quick brown fox jumps over the lazy dog."
      },
      "plainTranscript": "The quick brown fox jumps over the lazy dog.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 3801.03745800443,
      "inferenceMs": 3733.815499028424,
      "firstChunkMs": 2508.5268749971874,
      "audioSeconds": 3.4,
      "realTimeFactor": 1.1179521935307146,
      "audioSecondsPerSecond": 0.9105966807638766,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 2,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 3801.03745800443,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 0,
      "expectedSpeakerCount": 0,
      "speakerContent": {},
      "plainTranscript": "",
      "maximumConsecutiveRepeatedWords": 0,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1773.669999995036,
      "inferenceMs": 1714.5925830118358,
      "firstChunkMs": 1773.669415997574,
      "audioSeconds": 2,
      "realTimeFactor": 0.886834999997518,
      "audioSecondsPerSecond": 1.1664578628275766,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 1,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 1773.669999995036,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 7842.068499972811,
      "inferenceMs": 7757.6913320226595,
      "firstChunkMs": 2399.2004999890924,
      "audioSeconds": 11,
      "realTimeFactor": 0.7129153181793465,
      "audioSecondsPerSecond": 1.417947625035498,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 7842.068499972811,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 8110.002375004115,
      "inferenceMs": 8021.4134179695975,
      "firstChunkMs": 2366.461750003509,
      "audioSeconds": 11,
      "realTimeFactor": 0.7372729431821923,
      "audioSecondsPerSecond": 1.371329393814532,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 8110.002375004115,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 7946.097791980719,
      "inferenceMs": 7857.703332992969,
      "firstChunkMs": 2577.6626669976395,
      "audioSeconds": 11,
      "realTimeFactor": 0.7223725265437018,
      "audioSecondsPerSecond": 1.3999001405172855,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 7946.097791980719,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 7923.858249996556,
      "inferenceMs": 7837.128582003061,
      "firstChunkMs": 2457.5449169788044,
      "audioSeconds": 11,
      "realTimeFactor": 0.720350749999687,
      "audioSecondsPerSecond": 1.4035752871606648,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 7923.858249996556,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 7827.346417005174,
      "inferenceMs": 7739.859749999596,
      "firstChunkMs": 2502.3630829819012,
      "audioSeconds": 11,
      "realTimeFactor": 0.7115769470004704,
      "audioSecondsPerSecond": 1.4212143831159956,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 7827.346417005174,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12271616000,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 2924.03633298818,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 7782.563707994996,
      "inferenceMs": 7687.873791961465,
      "firstChunkMs": 2639.6545419993345,
      "audioSeconds": 11,
      "realTimeFactor": 0.7075057916359087,
      "audioSecondsPerSecond": 1.4308247374588452,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 10706.600040983176,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12051972096,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 2676.9912090094294,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 8098.800209001638,
      "inferenceMs": 8007.557833014289,
      "firstChunkMs": 2497.1548750181682,
      "audioSeconds": 11,
      "realTimeFactor": 0.7362545644546944,
      "audioSecondsPerSecond": 1.3737022234979306,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 10775.791418011067,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12046172160,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3409.955750015797,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2150.230042025214,
      "inferenceMs": 2061.5820820094086,
      "firstChunkMs": 958.1357920251321,
      "audioSeconds": 11,
      "realTimeFactor": 0.19547545836592856,
      "audioSecondsPerSecond": 5.335707996296894,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574049792,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 5560.185792041011,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " The quick brown fox jumps over the lazy dog."
      },
      "plainTranscript": "The quick brown fox jumps over the lazy dog.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 891.4839579956606,
      "inferenceMs": 832.5938750058413,
      "firstChunkMs": 683.3275000099093,
      "audioSeconds": 3.4,
      "realTimeFactor": 0.2622011641163708,
      "audioSecondsPerSecond": 4.083623603375831,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 2,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574361088,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 891.4839579956606,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 0,
      "expectedSpeakerCount": 0,
      "speakerContent": {},
      "plainTranscript": "",
      "maximumConsecutiveRepeatedWords": 0,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 439.8299169843085,
      "inferenceMs": 372.8274169843644,
      "firstChunkMs": 439.8286669747904,
      "audioSeconds": 2,
      "realTimeFactor": 0.21991495849215426,
      "audioSecondsPerSecond": 5.364412349759878,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 1,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651660800,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 439.8299169843085,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1922.1749159914907,
      "inferenceMs": 1834.18000000529,
      "firstChunkMs": 728.114875004394,
      "audioSeconds": 11,
      "realTimeFactor": 0.1747431741810446,
      "audioSecondsPerSecond": 5.997230369957297,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651660800,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1922.1749159914907,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1902.8923330188263,
      "inferenceMs": 1817.8275410027709,
      "firstChunkMs": 707.7809160109609,
      "audioSeconds": 11,
      "realTimeFactor": 0.17299021209262058,
      "audioSecondsPerSecond": 6.051179087061281,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651660800,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1902.8923330188263,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1819.4328749959823,
      "inferenceMs": 1737.9111670306884,
      "firstChunkMs": 677.3104160092771,
      "audioSeconds": 11,
      "realTimeFactor": 0.1654029886359984,
      "audioSecondsPerSecond": 6.329437435397847,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651660800,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1819.4328749959823,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1900.7825409935322,
      "inferenceMs": 1809.0135830279905,
      "firstChunkMs": 724.4135410001036,
      "audioSeconds": 11,
      "realTimeFactor": 0.17279841281759384,
      "audioSecondsPerSecond": 6.0806619160856785,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651660800,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1900.7825409935322,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1844.1454999847338,
      "inferenceMs": 1762.4733739648946,
      "firstChunkMs": 686.4049159921706,
      "audioSeconds": 11,
      "realTimeFactor": 0.16764959090770307,
      "audioSecondsPerSecond": 6.2412290378345885,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651660800,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1844.1454999847338,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098564608,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3210.933416004991,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2146.728209016146,
      "inferenceMs": 2057.923541025957,
      "firstChunkMs": 959.2745000263676,
      "audioSeconds": 11,
      "realTimeFactor": 0.19515710991055873,
      "audioSecondsPerSecond": 5.3451937259612965,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574049792,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 5357.661625021137,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 6655852544,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3510.1135000004433,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2440.0024579954334,
      "inferenceMs": 2338.6723330477253,
      "firstChunkMs": 1060.9730830183253,
      "audioSeconds": 11,
      "realTimeFactor": 0.2218184052723121,
      "audioSecondsPerSecond": 4.703523381432812,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574049792,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 5950.115957995877,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 7098253312,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 4344.67200000654,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 4087.297791003948,
      "inferenceMs": 3986.885500024073,
      "firstChunkMs": 2517.715666006552,
      "audioSeconds": 11,
      "realTimeFactor": 0.37157252645490435,
      "audioSecondsPerSecond": 2.759045876771124,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 10961682432,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 8431.969791010488,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " The quick brown fox jumps over the lazy dog."
      },
      "plainTranscript": "The quick brown fox jumps over the lazy dog.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1163.1682500010356,
      "inferenceMs": 1096.7590420041233,
      "firstChunkMs": 922.5431670201942,
      "audioSeconds": 3.4,
      "realTimeFactor": 0.34210830882383403,
      "audioSecondsPerSecond": 3.1000428259858537,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 2,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 10961993728,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1163.1682500010356,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 0,
      "expectedSpeakerCount": 0,
      "speakerContent": {},
      "plainTranscript": "",
      "maximumConsecutiveRepeatedWords": 0,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 482.7854579780251,
      "inferenceMs": 416.6835419891868,
      "firstChunkMs": 482.78429100173526,
      "audioSeconds": 2,
      "realTimeFactor": 0.24139272898901254,
      "audioSecondsPerSecond": 4.799805604157751,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 1,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11014127616,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 482.7854579780251,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2373.4093329985626,
      "inferenceMs": 2286.779748945264,
      "firstChunkMs": 839.7549579967745,
      "audioSeconds": 11,
      "realTimeFactor": 0.21576448481805113,
      "audioSecondsPerSecond": 4.810257745667702,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11014127616,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 2373.4093329985626,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2241.505000012694,
      "inferenceMs": 2157.112207933096,
      "firstChunkMs": 841.7572500184178,
      "audioSeconds": 11,
      "realTimeFactor": 0.20377318181933582,
      "audioSecondsPerSecond": 5.099410202003349,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11014127616,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 2241.505000012694,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2260.829917009687,
      "inferenceMs": 2182.2794989857357,
      "firstChunkMs": 809.7590419929475,
      "audioSeconds": 11,
      "realTimeFactor": 0.2055299924554261,
      "audioSecondsPerSecond": 5.040600896957746,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11014127616,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 2260.829917009687,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2247.5597089796793,
      "inferenceMs": 2170.0982919428498,
      "firstChunkMs": 816.5612089796923,
      "audioSeconds": 11,
      "realTimeFactor": 0.20432360990724358,
      "audioSecondsPerSecond": 5.068894824184161,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11014127616,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 2247.5597089796793,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2235.3540000040084,
      "inferenceMs": 2156.637709995266,
      "firstChunkMs": 815.4602500144392,
      "audioSeconds": 11,
      "realTimeFactor": 0.2032140000003644,
      "audioSecondsPerSecond": 5.10053216125213,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11014127616,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 2235.3540000040084,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760779264,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3713.4132500214037,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2584.176165983081,
      "inferenceMs": 2493.4970840113238,
      "firstChunkMs": 1146.0584579908755,
      "audioSeconds": 11,
      "realTimeFactor": 0.2349251059984619,
      "audioSecondsPerSecond": 4.411474980473667,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 10961682432,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 6297.5894160044845,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11760156672,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3655.770457990002,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2553.0080830212682,
      "inferenceMs": 2462.8444999980275,
      "firstChunkMs": 1130.6270000059158,
      "audioSeconds": 11,
      "realTimeFactor": 0.23209164391102438,
      "audioSecondsPerSecond": 4.466380236352238,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 10961682432,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 6208.77854101127,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 11806031872,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3130.51658400218,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 3927.2335419955198,
      "inferenceMs": 3826.493749016663,
      "firstChunkMs": 2709.327124990523,
      "audioSeconds": 11,
      "realTimeFactor": 0.3570212310905018,
      "audioSecondsPerSecond": 2.874694360294406,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574656000,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 7057.7501259977,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " The quick brown fox jumps over the lazy dog."
      },
      "plainTranscript": "The quick brown fox jumps over the lazy dog.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 909.1722090088297,
      "inferenceMs": 846.4546249888372,
      "firstChunkMs": 704.0809169993736,
      "audioSeconds": 3.4,
      "realTimeFactor": 0.2674035908849499,
      "audioSecondsPerSecond": 4.016753999122916,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 2,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574967296,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 909.1722090088297,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 0,
      "expectedSpeakerCount": 0,
      "speakerContent": {},
      "plainTranscript": "",
      "maximumConsecutiveRepeatedWords": 0,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 400.5893749999814,
      "inferenceMs": 344.07399996416643,
      "firstChunkMs": 400.5887089879252,
      "audioSeconds": 2,
      "realTimeFactor": 0.2002946874999907,
      "audioSecondsPerSecond": 5.812703081919267,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 1,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5652267008,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 400.5893749999814,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1819.0675410151016,
      "inferenceMs": 1734.7158330085222,
      "firstChunkMs": 689.7470000258181,
      "audioSeconds": 11,
      "realTimeFactor": 0.16536977645591833,
      "audioSecondsPerSecond": 6.3410962133911415,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5652267008,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1819.0675410151016,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1799.9294580076821,
      "inferenceMs": 1723.7946239474695,
      "firstChunkMs": 668.3869170083199,
      "audioSeconds": 11,
      "realTimeFactor": 0.1636299507279711,
      "audioSecondsPerSecond": 6.381270626549542,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5652267008,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1799.9294580076821,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1833.2953749923036,
      "inferenceMs": 1755.5740000098012,
      "firstChunkMs": 671.6254999919329,
      "audioSeconds": 11,
      "realTimeFactor": 0.16666321590839123,
      "audioSecondsPerSecond": 6.265756954670431,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5652267008,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1833.2953749923036,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1882.5023329991382,
      "inferenceMs": 1781.9757899851538,
      "firstChunkMs": 723.7510829872917,
      "audioSeconds": 11,
      "realTimeFactor": 0.17113657572719437,
      "audioSecondsPerSecond": 6.172923370688243,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5652267008,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1882.5023329991382,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 1842.1950420015492,
      "inferenceMs": 1766.086042043753,
      "firstChunkMs": 668.2899169973098,
      "audioSeconds": 11,
      "realTimeFactor": 0.16747227654559538,
      "audioSecondsPerSecond": 6.228462112338854,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5652267008,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 1842.1950420015492,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1007566848,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3193.998750008177,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2169.038250023732,
      "inferenceMs": 2076.778582006227,
      "firstChunkMs": 980.6940830021631,
      "audioSeconds": 11,
      "realTimeFactor": 0.1971852954567029,
      "audioSecondsPerSecond": 5.296664793881727,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574656000,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 5363.037000031909,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 759742464,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "maximumWordErrorRate": 0.25,
      "speakerCount": 1,
      "expectedSpeakerCount": 1,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country."
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": null,
      "maximumSpeakerWer": null,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3237.1377919916995,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 2152.8967500198632,
      "inferenceMs": 2062.7928740286734,
      "firstChunkMs": 965.7093750138301,
      "audioSeconds": 11,
      "realTimeFactor": 0.1957178863654421,
      "audioSecondsPerSecond": 5.332576110037064,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 4,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5574656000,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 5390.034542011563,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 761151488,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.0125,
      "maximumWordErrorRate": 0.35,
      "speakerCount": 2,
      "expectedSpeakerCount": 2,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
        "1": " The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]"
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. \n  And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": 0.0125,
      "maximumSpeakerWer": 0.35,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 10083.564041007776,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "linearQuantization": null,
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 21295.38541700458,
      "inferenceMs": 21199.360792001244,
      "firstChunkMs": 2372.6922500063665,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.5974402810262371,
      "audioSecondsPerSecond": 1.681389139499386,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 13,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 1,
      "totalMs": 31378.949458012357,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12466110464,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.0125,
      "maximumWordErrorRate": 0.35,
      "speakerCount": 2,
      "expectedSpeakerCount": 2,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
        "1": " The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]"
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. \n  And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": 0.0125,
      "maximumSpeakerWer": 0.35,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 2959.7039590007626,
      "provider": "PyTorch CPU",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "cpu/torch.float32": 1176
      },
      "acceleration": "cpu",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "linearQuantization": null,
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 27090.874540997902,
      "inferenceMs": 26994.56099900999,
      "firstChunkMs": 2649.6446250239387,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.7600322502778601,
      "audioSecondsPerSecond": 1.3204280299763804,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "cpu",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 13,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "cpuThreads": 4,
      "totalMs": 30050.578499998664,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "moderate",
      "sampledPeakHelperRssBytes": 12074631168,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.0125,
      "maximumWordErrorRate": 0.35,
      "speakerCount": 2,
      "expectedSpeakerCount": 2,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
        "1": " The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]"
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. \n  And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": 0.0125,
      "maximumSpeakerWer": 0.35,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3558.3164999843575,
      "provider": "Metal/MPS",
      "precision": "fp16",
      "parameterDtype": "torch.float16",
      "placement": {
        "mps:0/torch.float16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "linearQuantization": null,
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6867.568665998988,
      "inferenceMs": 6774.114666011883,
      "firstChunkMs": 1094.5539159874897,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.1926690723571107,
      "audioSecondsPerSecond": 5.261849962304354,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float16"
          }
        ]
      },
      "chunkCount": 13,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5650432000,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 10425.885165983345,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 6134284288,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.0125,
      "maximumWordErrorRate": 0.35,
      "speakerCount": 2,
      "expectedSpeakerCount": 2,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
        "1": " The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]"
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. \n  And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": 0.0125,
      "maximumSpeakerWer": 0.35,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3640.4949169955216,
      "provider": "Metal/MPS",
      "precision": "fp32",
      "parameterDtype": "torch.float32",
      "placement": {
        "mps:0/torch.float32": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "linearQuantization": null,
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 8277.208959014388,
      "inferenceMs": 8189.22300109989,
      "firstChunkMs": 1258.1250420189463,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.2322164144837548,
      "audioSecondsPerSecond": 4.352595477643315,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.float32"
          }
        ]
      },
      "chunkCount": 13,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 11012898816,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 11917.70387600991,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 12130680832,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.0125,
      "maximumWordErrorRate": 0.35,
      "speakerCount": 2,
      "expectedSpeakerCount": 2,
      "speakerContent": {
        "0": " And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country.",
        "1": " The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]"
      },
      "plainTranscript": "And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog. \n  And so, my fellow Americans, ask not what your country can do for you, ask what you can do for your country. \n  The quick brown fox jumps over the lazy dog. The quick brown fox jumps over the lazy dog.[Silence]",
      "maximumConsecutiveRepeatedWords": 1,
      "permutationInvariantSpeakerWer": 0.0125,
      "maximumSpeakerWer": 0.35,
      "scope": "Speaker-prefix native output; labels are not personal identities, no timestamps inferred"
    },
    "runtime": {
      "engine": "vibevoice-asr",
      "version": "1.0.0+16fb2cb",
      "transformersVersion": "4.57.6",
      "torchVersion": "2.10.0",
      "loadMs": 3212.288541981252,
      "provider": "Metal/MPS",
      "precision": "bf16",
      "parameterDtype": "torch.bfloat16",
      "placement": {
        "mps:0/torch.bfloat16": 1176
      },
      "acceleration": "partial",
      "accelerationDetails": "Acoustic/semantic encoders/connectors, Qwen decoder and logits on selected device; WAV decode/resample/tokenization/text decoding CPU",
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "linearQuantization": null,
      "chunkFrames": 22,
      "lookaheadFrames": 4,
      "sampleRate": 24000,
      "chunkSeconds": 2.933333333333333,
      "lookaheadSeconds": 0.5333333333333333,
      "attentionImplementation": "sdpa",
      "streaming": "Model chunk text emitted incrementally from complete WAV input",
      "durationMs": 6719.693749997532,
      "inferenceMs": 6619.256458070595,
      "firstChunkMs": 1047.2332909994293,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.18852045378822133,
      "audioSecondsPerSecond": 5.384951501091974,
      "activationPlacement": {
        "decoderLayer": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "logits": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "acousticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ],
        "semanticConnector": [
          {
            "device": "mps:0",
            "dtype": "torch.bfloat16"
          }
        ]
      },
      "chunkCount": 13,
      "temperature": 0,
      "maxNewTokensPerChunk": 256,
      "implicitMpsFallback": false,
      "driverAllocatedMemoryBytes": 5651038208,
      "memoryMeasurementScope": "Post-generation Metal driver allocation, not peak VRAM",
      "device": "mps",
      "cpuThreads": 4,
      "totalMs": 9931.982291978784,
      "artifactRevision": "4262d23d8a539a6530cf64fbd0b1751ef9a30853",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 768131072,
      "rssScope": "200ms helper process RSS sampling; not peak VRAM or validated minimum"
    }
  }
]

Something not quite right?

Tell us what happened with VibeVoice ASR Streaming. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.